On 07/05/06, Tim Wood <[EMAIL PROTECTED]> wrote:
Hi,
Here is a description of my test:
I have two thread groups which are identical and run in parallel. Each one
looks like this:
- Thread group - 15 threads, ramp 0, loop 300
-- Loop Controller - loop 5 times
---- request index.html
So that is repeated twice and run in parallel. If there is a much better way
that I should compose a test to try to put maximum load on a webserver, let
me know.
Unless the thread groups do something different, you might as well use
just the one:
- Thread group - 30 threads, ramp 0, loop 1500
-- request index.html
It would be better to use a ramp-up time > 0; say 10 or 15.
When I run the test from the command line, I get the worst results:
Generate Summary Results = 45000 in 86.3s = 521.5/s Avg: 49 Min: 1
Max: 48876 Err: 0 (0.00%)
Note that the maximum request rate is extremely high which can't be a good
sign.
It's not the maximum *rate*, it's the maximum *elapsed* time.
The first time I run through the GUI I get:
Generate Summary Results = 45000 in 82.8s = 543.8/s Avg: 27 Min: 1
Max: 9358 Err: 0 (0.00%)
The throughput for the first gui run is on par with the non-gui runs, but
does tend ot have a much lower max response time.
The second GUI run (clear results, and hit run again):
Generate Summary Results = 45000 in 48.5s = 928.1/s Avg: 16 Min: 1
Max: 2819 Err: 0 (0.00%)
Here my throughput almost doubles and my max response time drops again.
Repeating the test after that gives a small improvement up to about
1000req/s after 3 or 4 runs.
I've just tried doubling the length of the test (changed each thread group
from 300 to 600 loops), but get similar results. In this case, the test
takes about 160 seconds to complete in non gui (aka slow) mode. I also
noted that in 5 consecutive tests, the request with the maximum response
time (about 30-40 seconds!) always occured between 40 and 45 seconds into
the test.
Thanks for any suggestions or ideas.
Any error/warning messages in jmeter.log?
Or the console? (apart from garbage collection statistics).
Are you using keep-alive?
Perhaps the problem is related to connection handling, though I'm not
sure why it should improve on the second run.
Might be worth trying the same test with the HttpClient version of the
HTTP Sampler.
Also, try replacing the HTTP Sampler with a Java Request Sampler. This
will help show if there is a problem with JMeter itself.
S.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
